Scholarlink

<p>This paper presents a hybrid deep learning-based prediction model, i.e., a bidirectional long short-term memory-gated recurrent unit (BiLSTM-GRU) model to forecast the uncertain RER output powers and load demand data in the MG environment. The proposed BiLSTM-GRU model enhances the prediction accuracy by incorporating the strengths of both BiLSTM and GRU models, which capture the data with complex temporal patterns and interdependencies and reduce the complexity of the network structure. The prediction accuracy and performance efficiency of the proposed model have been compared with the different machine and deep learning models such as support vector machine (SVM), random forest regression (RF), gradient boosting machine (GBM), long-short term memory (LSTM), BiLSTM, GRU, and a hybrid LSTM-GRU with the help of various performance metrics. The performance evaluation has been tested with four distinct time horizons: 6&#xa0;h, 12&#xa0;h, 18&#xa0;h, and 24&#xa0;h ahead. The BiLSTM-GRU model outperformed when compared all other prediction models, achieving improved performance metric values, which are 0.000019 in MSE, 0.0020 in MAE, 0.0044 in RMSE, and 0.0246 in NRMSE for solar output power prediction. In the case of wind output power prediction, 0.00090 in MSE, 0.0215 in MAE, 0.0300 in RMSE, and 0.0763 in NRMSE values are obtained. Similarly, for the load demand data prediction, 0.00022 in MSE, 0.0124 in MAE, 0.0150 in RMSE, and 0.0206 in NRMSE are achieved. Furthermore, to validate the performance of the proposed model, a Diebold-Mariano (DM) test is performed. The DM results confirmed that the proposed model obtains the statistically significant improvements over benchmark models with average <i>p</i> values of 0.0288, 0.0410, and 0.027 (<i>p</i> &lt; 0.05) for solar output power, wind output power and load demand data, respectively. The computational efficiency of the proposed model has also demonstrated, which requires only 49&#xa0;s for training process and 0.3 s for testing process, and making it suitable for the real-time energy forecasting applications. The overall results analysis clearly shows the superior performance of the proposed BiLSTM-GRU model for the forecasting of RER output powers and load demand data with the highest accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A comparative study of machine learning and deep learning models for renewable energy and load demand forecasting in smart microgrids

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月3日

查看全文
  • K. Ravivarma,
  • B. Lokeshgupta,
  • Ramanjaneya Reddy Udumula,
  • Kiran Kumar Nallamekala

摘要

This paper presents a hybrid deep learning-based prediction model, i.e., a bidirectional long short-term memory-gated recurrent unit (BiLSTM-GRU) model to forecast the uncertain RER output powers and load demand data in the MG environment. The proposed BiLSTM-GRU model enhances the prediction accuracy by incorporating the strengths of both BiLSTM and GRU models, which capture the data with complex temporal patterns and interdependencies and reduce the complexity of the network structure. The prediction accuracy and performance efficiency of the proposed model have been compared with the different machine and deep learning models such as support vector machine (SVM), random forest regression (RF), gradient boosting machine (GBM), long-short term memory (LSTM), BiLSTM, GRU, and a hybrid LSTM-GRU with the help of various performance metrics. The performance evaluation has been tested with four distinct time horizons: 6 h, 12 h, 18 h, and 24 h ahead. The BiLSTM-GRU model outperformed when compared all other prediction models, achieving improved performance metric values, which are 0.000019 in MSE, 0.0020 in MAE, 0.0044 in RMSE, and 0.0246 in NRMSE for solar output power prediction. In the case of wind output power prediction, 0.00090 in MSE, 0.0215 in MAE, 0.0300 in RMSE, and 0.0763 in NRMSE values are obtained. Similarly, for the load demand data prediction, 0.00022 in MSE, 0.0124 in MAE, 0.0150 in RMSE, and 0.0206 in NRMSE are achieved. Furthermore, to validate the performance of the proposed model, a Diebold-Mariano (DM) test is performed. The DM results confirmed that the proposed model obtains the statistically significant improvements over benchmark models with average p values of 0.0288, 0.0410, and 0.027 (p < 0.05) for solar output power, wind output power and load demand data, respectively. The computational efficiency of the proposed model has also demonstrated, which requires only 49 s for training process and 0.3 s for testing process, and making it suitable for the real-time energy forecasting applications. The overall results analysis clearly shows the superior performance of the proposed BiLSTM-GRU model for the forecasting of RER output powers and load demand data with the highest accuracy.